Hipocrates Salud

Follow-up Engine is an AI-powered patient continuity engine that executes clinician-defined follow-up workflows, coordinates patient outreach, and schedules care without providing autonomous medical advice. The problem: a doctor orders an HbA1c and a control visit in three months. The patient never comes back, and nothing in the clinic notices. Orders without results, plans never completed and chronic patients who stop coming are lost silently, because no system tracks what should have happened. How it works: the clinic's EHR sends every saved consultation as an HL7 FHIR R4 bundle. A deterministic engine turns the doctor's orders and plan into dated expectations in four layers (orders and results, the doctor's own plan read by Meta Muse Spark with verbatim quotes, the patient's visit cadence, and condition protocols). When an order is overdue, the order-to-completion agent on Muse Spark takes ownership: it reads the authorized context, chooses the administrative action, contacts the patient on WhatsApp (simulated in the demo), understands text, voice notes and photos of lab reports, offers slots computed by code, books only on an explicit yes, and escalates to the clinic team when the patient does not answer. The agent never marks an order done: the engine verifies the FHIR Appointment or Observation and closes it. Anything clinical, like a rising HbA1c, goes to the doctor as a finding with citations, never to the patient. Agent acts, engine verifies, doctor decides. The demo is presented by Dr. Luis Pinerua, the team's medical lead and co-founder, from the clinician's side: the problem he sees in practice, the orders he writes, and the decisions that stay with him. Evidence: practicing physicians confirmed 42 plan items read by Muse Spark (0 corrected, 0 rejected); 0 of 348 hallucinated quotes in extraction evals; in a 90-day synthetic simulation, orders fulfilled went from 50.6% with fixed reminders to 74.6% with the agent. 3,109 tests pass on Postgres 18. What you see is a simulation: every patient, message and result is synthetic, and the WhatsApp channel is simulated. Next, we will deploy the engine inside Hipócrates, our EHR already used by clinics in Latin America, which connects to it over FHIR without changes. That gives us real clinics, real orders and physicians to validate it with quickly